Colloidal nanocrystals (NCs) are crucial in various applications due to their unique physicochemical properties. Achieving precise control over their morphology is essential for tailoring these properties. However, conventional synthesis methods often rely on time-consuming trial-and-error approaches and labor-intensive characterization. This research aims to develop a robotic platform to automate the synthesis and characterization processes, leading to efficient morphology control. The automation is designed to accelerate the discovery and synthesis of NCs with desired morphologies. The platform's effectiveness is demonstrated using gold nanocrystals (known for strong visible-light absorption) and double-perovskite nanocrystals (known for photoluminescence) as model systems, representing both established and emerging research areas in nanomaterials. The combination of automation, in situ characterization, and machine learning is expected to significantly improve the efficiency and precision of nanocrystal synthesis, ultimately advancing the field of materials science.

Robotic synthesis has gained momentum in organic materials synthesis, with platforms automating synthesis, characterization, and database generation, often integrated with artificial intelligence (AI) for retrosynthesis. However, the application to inorganic materials, particularly colloidal NCs, is less explored. While some robotic platforms exist for inorganic materials synthesis, the integration of data mining from literature to guide robotic search for targeted inorganic materials, especially NCs, has been limited. This study addresses this gap by incorporating data-driven initial hypotheses, robot-assisted synthesis, and experimental databases within a unified robotic platform. This framework combines existing literature data with experimental data generated by the robotic system, using machine learning to identify correlations between synthesis parameters and resulting morphologies.

The robotic platform consists of three main stages: data mining, controllable synthesis, and inverse design. Data mining uses an automated literature recommendation system to extract key synthesis parameters from publications for both gold and double-perovskite NCs. For gold NCs, the system analyzed 1300 studies to determine the frequency distribution of surfactant (CTAB) and other agent concentrations. For double-perovskite NCs, potential solvents and surfactants were identified from related literature. Controllable synthesis utilizes high-throughput experiments (orthogonal design, single-factor, and double-factor experiments) with automated synthesis modules (pipettes, storage, synthesis platform) and in situ characterization modules (spectrometer, color-sensitive camera, light sources). Two collaborative robots assist in microplate transport and equipment servicing. Robotic Execution Excel files and a simulated operation system manage the automated processes. The acquired data (absorption spectra, RGB values from images, TEM, SEM) are used to train machine learning models (using SISSO algorithm) to identify correlations between structure-directing agents (SDAs) and NC morphologies (LSPR, aspect ratio, size). Finally, inverse design uses these trained ML models to predict the SDAs needed to achieve targeted NC morphologies. The methodology involved extensive experimentation, including orthogonal experiments for initial parameter screening, single-factor experiments to analyze individual parameter effects, double-factor and triple-factor experiments to study interactions between parameters, and finally, inverse design experiments to validate the predictive capabilities of the ML models. The platform generated a comprehensive database of over 3300 samples.

The robotic platform successfully synthesized both gold and double-perovskite NCs with controlled morphologies. For gold NCs, the platform identified CTAB, AgNO3, and HCl as key SDAs influencing the longitudinal surface plasmon resonance (LSPR) and aspect ratio. Machine learning models revealed synergistic effects between CTAB and AgNO3 in controlling LSPR and a linear relationship between LSPR and aspect ratio. For double-perovskite NCs, PVP and BiCl3 were identified as crucial SDAs influencing crystal size. Machine learning models successfully correlated the normalized absorption at 400 nm with crystal size. The platform also generated a robust database from the in situ characterized data (ultraviolet-visible-near-infrared absorption spectra, RGB values) and the ex situ validation data (TEM and SEM images). This database allowed for the development of ML models that accurately predicted the relationship between synthesis parameters and the resulting morphologies. The accuracy of these models was validated through a series of experiments, including those targeting specific morphologies based on predictions (inverse design). Inverse design experiments successfully synthesized gold NCs with a target aspect ratio and double-perovskite NCs with targeted sizes, demonstrating the platform's ability to accurately predict and achieve desired morphologies.

The findings address the limitations of traditional nanocrystal synthesis by demonstrating a robust and efficient robotic platform. This platform effectively integrates data mining, automated synthesis and characterization, and machine learning for precise morphology control. The successful synthesis of gold and double-perovskite NCs with desired morphologies validates the platform's effectiveness. The use of machine learning allows for the prediction of synthesis parameters necessary to obtain specific morphologies, enabling a more efficient and targeted approach to nanomaterial synthesis. The high-throughput nature of the platform significantly reduces the time and resources needed for nanomaterial research, paving the way for faster discovery and development of advanced materials. The integration of in situ characterization and machine learning enhances the efficiency and precision of the process, enabling the rapid optimization of synthesis parameters.

This study successfully demonstrates a robotic platform for the synthesis of colloidal nanocrystals with controlled morphologies. This platform integrates data mining, automated synthesis and in situ characterization, and machine learning for inverse design. The successful synthesis of gold and double-perovskite NCs with specific morphologies highlights the platform's potential for accelerating nanomaterial research and development. Future research could focus on expanding the platform's capabilities to a broader range of nanomaterials and exploring more sophisticated machine learning models for enhanced predictive accuracy.

The current platform is focused on two specific types of nanocrystals. While the methodology is broadly applicable, further validation is needed to confirm its effectiveness for other nanomaterials. The accuracy of the machine learning models relies heavily on the quality and quantity of the data used for training. The platform's dependence on the quality of the literature data used for initial parameter selection could also be a limitation, as the accuracy of the initial parameters could influence the overall efficiency of the system. The reliance on currently available literature could also introduce bias based on the historical trends in the research.